Performance change detection and recovery in inferential control loops

In most industrial environments, soft sensors are crucial for estimating variables that are hard to measure. It is critical that these variables be estimated quickly and accurately because they are closely linked to plant safety and profit. Therefore, performance drift in the soft sensors cannot be ignored. This drift is a direct result of a mismatch between the current behaviour of the plant and the soft sensor. This deviation is often defined as plant–model mismatch (PMM). Although PMM is unavoidable, it is possible to detect and minimize it. For a large number of input/output processes, re‐identifying the models is a time‐consuming and expensive proposition. This paper introduces an approach to assess the impact of PMM and adjust the control signal accordingly, without altering the soft sensor. Initially, the performance change index (PCI) is defined to measure the performance variation induced by PMM by using coprime factorization and small‐gain theory. Then, according to the PCI and the Youla–Kučera parameterization, the modified rules of the controller are given. When the PMM is detected to be at an unacceptable level, a compensation signal is generated to counteract the PMM. A continuous stirred tank reactor (CSTR) and a three‐tank system are used to validate the efficacy of the proposed approach. The results show that when the performance deteriorates to an unacceptable level, the compensation signal mechanism will be triggered to compensate for the interference of the PMM signal, and the performance restored to the normal level.